We will develop a new class of algorithms, of value in many medical image processing tasks. They will be designed around a synthesis of partial differential equations (PDE's) and stochastic methods. Eventually, the PDE's themselves will be modeled stochastically: That is, we will use a partial differential approach without the partial differential equations. These new algorithms will provide a fundamental substrate for new methods in segmentation, registration, and shape analysis.